Is every claim in the answer grounded in the retrieved context? Detects hallucination.
Answer Relevance
0.76Good
Does the answer actually address the question asked? Low = off-topic or partial.
Context Recall
0.68Needs Work
Did the retriever fetch all the chunks needed to answer? Low = missing key info.
Context Precision
0.71Good
Were the retrieved chunks relevant? Low = noisy, irrelevant context injected.
Controls
Presets
Faithfulness
Faithfulness0.82
Answer Relevance
Answer0.76
Context Recall
Context0.68
Context Precision
Context0.71
RAGAS evaluates RAG pipelines end-to-end using 4 metrics. The overall score is their geometric mean - one weak metric tanks everything. Try the "Hallucinating LLM" preset to see how faithfulness collapses the score.
RAG Evaluation with RAGAS - Interactive Visualization
RAGAS evaluates RAG pipelines end-to-end using four metrics: faithfulness (is the answer grounded in context?), answer relevance (does it address the question?), context recall (was all needed information retrieved?), and context precision (was the retrieved context relevant?). The overall RAGAS score is the geometric mean of all four - one weak metric tanks the whole pipeline.
Faithfulness detects hallucination by checking if every claim in the answer is grounded in retrieved context
Context recall catches retrieval failures where key information was never fetched
Context precision catches noisy retrieval where irrelevant chunks pollute the LLM prompt
The geometric mean scoring means you cannot compensate a weak metric with strong ones elsewhere
Part of the EngineersOfAI Interactive 3D - free interactive visualizations covering every major concept in machine learning and AI engineering. Hover any element for a plain-English explanation. No code required.